What a datum actually is

The reference everything else is measured from, and how it is chosen.

A datum is the specific feature on a part, a surface, an edge, an axis, that every other dimension on the drawing is deliberately measured from, and it is chosen because it is the feature that matters to how the part functions once it is assembled and put to work.

Winchester's Exactly traces precision from the first accurate cannon bore to the machines that make chips, treating tolerance as a genuine historical force, and reading it made clear that almost every story of a precision failure in the book eventually traces back to the same root question this article is about, what a measurement was actually being taken from.

Twenty millimetres from where

A dimension on a drawing, "twenty millimetres," means nothing on its own until someone answers twenty millimetres from where. A datum answers that by fixing the starting point every related dimension is measured from, so a hole positioned "twenty millimetres from the datum edge" is unambiguous in a way "twenty millimetres from somewhere on the part" never could be. Most real parts need several datums arranged in a deliberate hierarchy, a primary, a secondary and a tertiary, echoing the three-plane locating scheme an earlier set in this project covered for physically holding a part still. A drawing's datum scheme and a fixture's locating scheme describe the same idea from two directions.

The primary datum fixes the part against its main registration surface. The secondary and tertiary then pin down what is left, one stopping any remaining rotation and the other any remaining sideways slide, so the set locates the part completely without duplicating a constraint the primary has already handled.

Heights above sea level

A mountain's height is quoted above sea level, never from the base of the hill it happens to sit on, because sea level is one agreed reference that lets two mountains rising from very different ground be compared on the same scale. It was chosen for consistency across the whole world, and ease of reaching it with a tape measure had nothing to do with it. Even sea level has to be defined with care, since tides, currents and the earth's uneven gravity move the ocean's surface constantly, so surveyors work from a mean sea level averaged over many years at an agreed reference station. A datum on an engineering drawing plays the same role at a much smaller scale, so that every dimension referencing it means the same, comparable thing whichever feature someone is looking at.

The surface the part registers against

The instinctive choice for a datum is whatever surface is flattest, largest, or easiest to set the part down on. Sometimes that is right, and often it is wrong, because the feature that matters is the one the part will register against once it is assembled: a mounting face bolted to another component, a bore a shaft will pass through, a locating pin that positions the part relative to its neighbours.

A drawing built on the convenient surface is internally consistent and functionally misleading. A part can pass every check against its own datum and still fail to line up with the parts around it, a mismatch that often appears only the first time real parts are put together, because inspection only ever compared the part with itself. Asking what the part registers against once installed, before picking up a pencil, usually settles the choice.

A datum has to be touchable

A datum is useful only if it can be found again later, by an inspector with a gauge, by a fixture holding the part for machining, and by the mating part it will assemble against. That is why a datum reference is always tied to a real physical feature of the part. A theoretical centre point buried inside a solid part may be mathematically convenient, yet no gauge can probe it and no fixture can rest against it, so a usable scheme builds the reference from accessible surfaces, a flat face, a round bore, an edge, that anyone measuring the part later can contact and repeat.

Reading a drawing well starts with asking what each dimension is measured from and why that feature was chosen, since a beautifully toleranced dimension from the wrong datum is precision spent on the wrong problem. The next article in this set follows what happens once a real measurement is taken from a well-chosen datum, starting with the uncomfortable fact that measuring the same part twice, with the same instrument, rarely gives the same number both times.